Fabio/epub_corrector
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
No description
1 commit 1 branch 0 tags 221 MiB
Python 90.7%
HTML 6.8%
Shell 2.5%
Find a file
Fabio 6c27e8bddb Add EPUB corrector runtime, web UI, tokenizer and RKNN model
2026-10-30 16:08:03 +01:00
assets/tokenizer Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
epub-corrector-ui Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
epub-corrector-ui-app Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
rknn Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
.gitattributes Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
.gitignore Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
apply_approved_epub_changes.py Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
config.example.json Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
epub_npu_dry_run_grouped_v2.py Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
generate_review_html.py Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
guida_orange_pi_epub_npu.md Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
README.md Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00
requirements-rknn.txt Add EPUB corrector runtime, web UI, tokenizer and RKNN model 2026-10-30 16:08:03 +01:00

README.md

Fase 2: dataset per classificatore italiano

Genera coppie bilanciate [frase A, frase B], dove la label indica se B è la versione corretta.

python make_training_pairs.py ~/epub-corrector/input/a.epub --pairs 20000 --output pairs.jsonl
python check_dataset.py pairs.jsonl
python split_dataset.py pairs.jsonl --outdir data

Nota: usare un solo libro rischia overfitting stilistico. Questa è una prova di fattibilità; per il modello finale servono più fonti italiane pulite e un test manuale separato.